T

Lead Research Engineer, Large-Scale Agentic RL Training

Thomson Reuters·Zug·19.11.2026

 thomsonreuters.wd5.myworkdayjobs.com

 
 
full_time80–100%
Job written in
English
Location
Zug
Work type
Hybrid
Type
Full-time

 

Thomson Reuters Labs is looking for a Lead Research Engineer to shape the infrastructure that trains agentic large language models (LLMs) with reinforcement learning. The team builds frontier‑scale systems that handle Mixture‑of‑Experts models larger than 300 billion parameters, running asynchronous RL loops, long‑context training, and tool‑driven agent harnesses. The role sits at the intersection of research and engineering, ensuring that new algorithms, rewards and environments can be evaluated quickly and reliably. On a daily basis you will own the technical direction of the RL training stack, deciding how it evolves for bigger models, longer contexts and more complex workloads. You will profile the entire RL pipeline, from rollout generation through reward computation to weight synchronization, and cut step time and GPU idle periods. Designing asynchronous, fault‑tolerant pipelines that let models interact with tools during training is a core task, as is debugging numerical, distributed and performance issues across the training framework, inference engine and model code. You will also partner closely with Research Scientists to enable rapid experimentation and set engineering standards for code quality, testing and reproducibility while mentoring junior engineers. The position requires an MSc or PhD in computer science, machine learning or a related discipline, or equivalent industry experience, plus at least three years of professional work in ML or research engineering on large‑scale model training. You must have deep experience training massive models on multi‑node GPU clusters using PyTorch (advanced) and Megatron‑LM/Core, as well as familiarity with FSDP/DTensor or DeepSpeed. A strong grasp of parallelism strategies, including tensor, pipeline, context and expert parallelism, and their memory and communication trade‑offs is essential, especially for MoE and long‑context models. Hands‑on work with LLM inference engines such as vLLM or SGLang, GPU profiling, NCCL communication, CUDA graphs and kernel bottlenecks is also required, together with solid software‑engineering practices in large open‑source codebases. Preferred experience includes building or extending RL post‑training frameworks (e.g., NeMo‑RL, verl, OpenRLHF, TRL), especially asynchronous designs, and training or serving models over 100 billion parameters with 100k+ token contexts. Familiarity with distributed runtimes and orchestration tools like Ray, Kubernetes or Slurm on cloud platforms (AWS, Azure, Google Cloud) is a plus, as is work on numerical‑parity testing between training and inference. Contributions to open‑source projects such as vLLM, Megatron or PyTorch, and publications at venues like MLSys, NeurIPS or OSDI, are also valued. The role offers a hybrid work model with flexible "Flex My Way" policies, allowing up to eight weeks of remote work per year and generous work‑life balance benefits. Employees receive comprehensive benefits, including flexible vacation, mental‑health days, tuition reimbursement and retirement savings, plus a culture that emphasizes inclusion, continuous learning and social impact through volunteer days and ESG initiatives. What the role asks for: - MSc or PhD in computer science, ML or related field - 3+ years professional ML or research engineering experience - Advanced PyTorch expertise on multi‑node GPU clusters - Experience with Megatron‑LM / Megatron‑Core - Hands‑on with FSDP, DTensor or DeepSpeed - Strong knowledge of tensor, pipeline, context and expert parallelism - Experience with LLM inference engines (vLLM, SGLang) - GPU profiling and optimisation skills (NCCL, CUDA graphs, kernels)

 

 

 

 

 

Kostenlos

Engineer: Lebenslauf-Vorlage

Bewirb dich mit einem Lebenslauf im Schweizer Aufbau — mit Beispieltext und den Anforderungen, die Engineer-Inserate am häufigsten nennen.

Lebenslauf-Vorlage Schweiz ansehen

Engineer: was der Markt gerade verlangt

Engineer-Stellen gehören zu den regelmässig ausgeschriebenen Berufen auf SwissJobs.app.

Weitersuchen

Ähnliche Jobs per E-Mail erhalten